Disk-IO chart not correct

[udo]

Hi,
just wonder about the performance-charts. They can't be right.
At the attached screenshot i transfer an 30GB-sql file to an openvz-vm:
I see the network traffic but no disk-io. After that i start an import - and see only disk-read!


On the host i see also writes (only this VM is running):

root@proxtest:~# iostat -dm 5 sda
Linux 2.6.32-6-pve (proxtest)   11/15/2011      _x86_64_        (4 CPU)

Device:            tps    MB_read/s    MB_wrtn/s    MB_read    MB_wrtn
sda            1374.53         0.75         6.81       4816      43542

Device:            tps    MB_read/s    MB_wrtn/s    MB_read    MB_wrtn
sda            1793.20         1.76         8.29          8         41

Device:            tps    MB_read/s    MB_wrtn/s    MB_read    MB_wrtn
sda            2055.40         1.73         8.15          8         40

Device:            tps    MB_read/s    MB_wrtn/s    MB_read    MB_wrtn
sda             759.80         1.73         8.64          8         43

Also show the CPU up to 1450% of 2CPUs (the host has only 4 cores... so 1450% is very much).

Udo

 

[e100]

My system thought I was writing at 300PB!


As soon as that erroneous spike went by and the graph re-adjusted it looked normal.
This happened when I had the disk cache=writeback, it only wrote at about 2GB/sec but the graph shows 300PB/sec.

 

[dietmar]

I see the network traffic but no disk-io.

I have not really found an accurate way to measure disk io for OpenVZ. I currently use the blkio cgroup, but that seems to return nonsense. I still hope that gets fixed in later kernels.

 

[e100]

I can reproduce my 300PB graph by doing a live migration of a KVM VM while doing lots of disk IO.

 

[udo]

...
Also show the CPU up to 1450% of 2CPUs (the host has only 4 cores... so 1450% is very much).

Hi,
just reconice that the cpu-value simply factor 100 to high:

 

[dietmar]

Hi,
just reconice that the cpu-value simply factor 100 to high:

just fixed in git.